micro-RNA Vesicles

Устанавливаем необходимые библиотки

library(tidyverse)
library(DESeq2)
library(pheatmap)
library(RColorBrewer)
library(clusterProfiler)
library(biomaRt)
library(org.Hs.eg.db)
library(EnhancedVolcano)
library(GenomicRanges)
library(msigdbr)
library(multiMiR)
library(miRBaseConverter)
library(enrichplot)
library(vsn)
library(rvest)
library(patchwork)
library(dbplyr)

Импортируем данные

setwd("/Users/dariakilina/GitHub/Micro-RNA_cardiology")
coldata <- read_tsv("data/phenotableV.tsv", show_col_types = FALSE)
coldata$type <- as.factor(coldata$type)
coldata$patient <- as.factor(coldata$patient)
coldata$condition <- as.factor(coldata$condition)
coldata <- as.data.frame(coldata)
rownames(coldata) <- coldata$sample
coldata

counts <- read.csv("data/miR.Counts.csv", header = TRUE, sep = ",")
counts <- column_to_rownames(counts, var = "miRNA")
#counts <- round(counts) # если используем нормализованные данные
head(counts)

colnames(counts) <- gsub("^X", "", colnames(counts))
common_samples <- intersect(colnames(counts), coldata$sample)

counts <- counts[, c(counts$miRNA, common_samples)]  
counts <- counts[, rownames(coldata)] #ранжирую по колонки в counts так же как и названия строк в coldata
head(counts)

anno <- read.csv("data/annotation.report.csv", header = TRUE, sep = ",")
anno$Sample.name.s. <- gsub("-", ".", anno$Sample.name.s.)
anno <- anno[, -c(2:5, 7, 15)]


common_samples <- intersect(anno$Sample.name.s., coldata$sample)

anno <- anno[anno$Sample.name.s. %in% common_samples, ]
anno <- anno[match(rownames(coldata), anno$Sample.name.s.), ] #ранжирую по колонки в counts так же как и названия строк в coldata
anno

Весь датасет

anno_long <- anno %>%
  pivot_longer(cols = -Sample.name.s., names_to = "RNA_Type", values_to = "Count")

plt <- ggplot(anno_long, aes(x = Sample.name.s., y = Count, fill = RNA_Type)) +
  geom_bar(stat = "identity") +
  theme_minimal() +
  labs(x = "Sample", y = "Read Count") +
  theme(axis.text.x = element_text(angle = 45, hjust = 1)) +
  scale_fill_brewer(palette = "Set3")  # Красивые цвета

print(plt)
ggsave("./pictures/barplot_alldataset_no_normalised.tiff", plot = plt, width = 8, height = 6, dpi = 300,  bg = "white")

anno_long <- anno %>%
  rowwise() %>%
  mutate(across(-Sample.name.s., ~ . / sum(c_across(-Sample.name.s.)))) %>% 
  ungroup() %>%
  pivot_longer(cols = -Sample.name.s., names_to = "RNA_Type", values_to = "Proportion")

plt <- ggplot(anno_long, aes(x = Sample.name.s., y = Proportion, fill = RNA_Type)) +
  geom_bar(stat = "identity") +
  theme_minimal() +
  labs(x = "Sample", y = "Proportion") +
  theme(axis.text.x = element_text(angle = 45, hjust = 1)) +
  scale_fill_brewer(palette = "Set3")

plt
ggsave("./pictures/barplot_alldataset_normalised.tiff", plot = plt, width = 8, height = 6, dpi = 300, bg = "white")

coldata$condition <- relevel(coldata$condition, ref = "before")
modelMatrix <- model.matrix(~type*condition + patient, data = coldata)
modelMatrix

                    (Intercept) type150 conditionafter patient17 patient29 type150:conditionafter
29.1p16_S39_R1_001            1       0              0         0         1                      0
15.1p16_S33_R1_001            1       0              0         0         0                      0
17.1p16_S35_R1_001            1       0              0         1         0                      0
29.1p150_S40_R1_001           1       1              0         0         1                      0
15.1p150_S34_R1_001           1       1              0         0         0                      0
17.1p150_S36_R1_001           1       1              0         1         0                      0
29.7p16_S41_R1_001            1       0              1         0         1                      0
15.7p16_S31_R1_001            1       0              1         0         0                      0
17.7p16_S37_R1_001            1       0              1         1         0                      0
15.7p150_S32_R1_001           1       1              1         0         0                      1
29.7p150_S42_R1_001           1       1              1         0         1                      1
17.7p150_S38_R1_001           1       1              1         1         0                      1
attr(,"assign")
[1] 0 1 2 3 3 4
attr(,"contrasts")
attr(,"contrasts")$type
[1] "contr.treatment"

attr(,"contrasts")$condition
[1] "contr.treatment"

attr(,"contrasts")$patient
[1] "contr.treatment"

dds <- DESeqDataSetFromMatrix(countData = counts, 
                              colData = coldata, 
                              design = ~type*condition + patient)

converting counts to integer mode

dds$condition <- relevel(dds$condition, ref = "before")
dds

class: DESeqDataSet 
dim: 913 12 
metadata(1): version
assays(1): counts
rownames(913): Hsa-Let-7-P1a_3p* Hsa-Let-7-P1a_5p/P2a1_5p/P2a2_5p ... Hsa-Mir-9851_3p Hsa-Mir-9851_5p*
rowData names(0):
colnames(12): 29.1p16_S39_R1_001 15.1p16_S33_R1_001 ... 29.7p150_S42_R1_001 17.7p150_S38_R1_001
colData names(4): sample condition type patient

dim(dds)

[1] 913  12

smallestGroupSize <- 3
keep <- rowSums(counts(dds) >= 10) >= smallestGroupSize
dds <- dds[keep,]
dim(dds)

[1] 292  12

Run Differential Expression Analysis for all dataset

dds <- DESeq(dds, fitType = "parametric")

estimating size factors
estimating dispersions
gene-wise dispersion estimates
mean-dispersion relationship
final dispersion estimates
fitting model and testing

dds

class: DESeqDataSet 
dim: 292 12 
metadata(1): version
assays(4): counts mu H cooks
rownames(292): Hsa-Let-7-P1a_5p/P2a1_5p/P2a2_5p Hsa-Let-7-P1b_5p ... Hsa-Mir-96-P2_5p Hsa-Mir-96-P3_5p
rowData names(38): baseMean baseVar ... deviance maxCooks
colnames(12): 29.1p16_S39_R1_001 15.1p16_S33_R1_001 ... 29.7p150_S42_R1_001 17.7p150_S38_R1_001
colData names(5): sample condition type patient sizeFactor

plotDispEsts(dds)

raw_counts <- counts(dds, normalized = FALSE)
normalized_counts <- counts(dds, normalized = TRUE)

df <- data.frame(
  Sample = rep(colnames(dds), 2),
  Counts = c(colSums(raw_counts), colSums(normalized_counts)),
  Type = rep(c("Raw", "Normalized"), each = ncol(dds))
)

plt <- ggplot(df, aes(x = Sample, y = Counts, fill = Type)) +
  geom_bar(stat = "identity", position = "dodge") +
  theme_minimal() +
  labs(title = "Counts before and after normalization", x = "Sample", y = "Total Counts") +
  theme(axis.text.x = element_text(angle = 45, hjust = 1))

plt
ggsave("./pictures/Counts before and after normalization.tiff", plot = plt, width = 8, height = 6, dpi = 300, bg = "white")

rlog трансформация

rlt <- rlog(dds)  #rlog Transformation
meanSdPlot(assay(rlt)) 

vsd <- varianceStabilizingTransformation(dds, blind=FALSE) 
meanSdPlot(assay(vsd)) #показывает, как изменяется стандартное отклонение в зависимости от среднего значения экспрессии

PCA plot

pcaData <- plotPCA(rlt, intgroup=c("condition", "type", "patient"), returnData = TRUE)

using ntop=500 top features by variance

percentVar <- round(100 * attr(pcaData, "percentVar"))

pcaData$condition_type <- paste(pcaData$condition, pcaData$type, sep = "_")

ggplot(pcaData, aes(PC1, PC2, shape = patient, color = condition_type)) +
  geom_point(size = 3) +
  xlab(paste0("PC1: ", percentVar[1], "%")) +
  ylab(paste0("PC2: ", percentVar[2], "%")) + 
  coord_fixed() +
  theme_bw() +
  ggtitle("PCA plot for all dataset before removing donor effect")+
  scale_color_brewer(palette = "Set2")

Plot a heatmap of 50 most expressed genes Этот heatmap отражает уровни экспрессии генов, а не разницу между группами. Цвета не означают up- или down-регуляцию в сравнении с контрольной группой, потому что heatmap показывает абсолютные значения экспрессии, а не fold change!

Plot of the distance between samples heatmap Расчет расстояний между образцами • Обычно используется евклидово расстояние (по умолчанию в DESeq2). • Оно вычисляется по нормализованным данным экспрессии (rlog() или vst()). • Чем меньше расстояние — тем более похожи образцы.

sampleDists <- dist(t(assay(rlt)))
sampleDistMatrix <- as.matrix(sampleDists)
rownames(sampleDistMatrix) <- paste(rlt$condition, rlt$type, sep="_type")
colnames(sampleDistMatrix) <- paste(rlt$condition, rlt$type, sep="_type")
colors <- colorRampPalette(rev(brewer.pal(9, "Blues")) )(255)

plt <- pheatmap(sampleDistMatrix,
         clustering_distance_rows = "euclidean",
         clustering_distance_cols = "euclidean",
         color = colors)

plt

ggsave("./pictures/Plot of the distance between samples heatmap.tiff", plot = plt, width = 8, height = 6, dpi = 300, bg = "white")

Только везикулы типа 150

coldata_150 <- coldata[coldata$type == 150, ]
rownames(coldata_150) <- coldata_150$sample
coldata_150

common_samples_150 <- intersect(colnames(counts), coldata_150$samples)

counts_150 <- counts[, c(counts$miRNA, common_samples)]  
counts_150 <- counts_150[, rownames(coldata_150)] #ранжирую по колонки в counts так же как и названия строк в coldata_150
head(counts_150)

coldata_150$condition <- relevel(factor(coldata_150$condition), ref = "before")
modelMatrix <- model.matrix(~ 0 + patient + condition , coldata)
modelMatrix

                    patient15 patient17 patient29 conditionafter
29.1p16_S39_R1_001          0         0         1              0
15.1p16_S33_R1_001          1         0         0              0
17.1p16_S35_R1_001          0         1         0              0
29.1p150_S40_R1_001         0         0         1              0
15.1p150_S34_R1_001         1         0         0              0
17.1p150_S36_R1_001         0         1         0              0
29.7p16_S41_R1_001          0         0         1              1
15.7p16_S31_R1_001          1         0         0              1
17.7p16_S37_R1_001          0         1         0              1
15.7p150_S32_R1_001         1         0         0              1
29.7p150_S42_R1_001         0         0         1              1
17.7p150_S38_R1_001         0         1         0              1
attr(,"assign")
[1] 1 1 1 2
attr(,"contrasts")
attr(,"contrasts")$patient
[1] "contr.treatment"

attr(,"contrasts")$condition
[1] "contr.treatment"

Создаем DESeqDataSet из матрицы каунтов

dds_150 <- DESeqDataSetFromMatrix(countData = counts_150, 
                              colData = coldata_150, 
                              design = ~ 0 + patient + condition)

converting counts to integer mode

dds_150$condition <- relevel(dds_150$condition, ref = "before")
dds_150

class: DESeqDataSet 
dim: 913 6 
metadata(1): version
assays(1): counts
rownames(913): Hsa-Let-7-P1a_3p* Hsa-Let-7-P1a_5p/P2a1_5p/P2a2_5p ... Hsa-Mir-9851_3p Hsa-Mir-9851_5p*
rowData names(0):
colnames(6): 29.1p150_S40_R1_001 15.1p150_S34_R1_001 ... 29.7p150_S42_R1_001 17.7p150_S38_R1_001
colData names(4): sample condition type patient

Фильтрация

dim(dds_150)

[1] 913   6

smallestGroupSize <- 3
keep <- rowSums(counts(dds_150) >= 10) >= smallestGroupSize
dds_150 <- dds_150[keep,]
dim(dds_150)

[1] 240   6

Run Differential Expression Analysis for 150 type

dds_150 <- DESeq(dds_150, fitType = "parametric")

using pre-existing size factors
estimating dispersions
found already estimated dispersions, replacing these
gene-wise dispersion estimates
mean-dispersion relationship
final dispersion estimates
fitting model and testing

plotDispEsts(dds_150)

res_150 <- results(dds_150, contrast=c("condition", "before", "after"))
res_150

log2 fold change (MLE): condition before vs after 
Wald test p-value: condition before vs after 
DataFrame with 240 rows and 6 columns
                                   baseMean log2FoldChange     lfcSE      stat    pvalue      padj
                                  <numeric>      <numeric> <numeric> <numeric> <numeric> <numeric>
Hsa-Let-7-P1a_5p/P2a1_5p/P2a2_5p  9637.8166      0.3135594  0.384984  0.814474 0.4153737  0.951261
Hsa-Let-7-P1b_5p                   578.3774      0.7372962  0.710691  1.037435 0.2995332  0.930944
Hsa-Let-7-P1c_5p                   390.5366      1.3620176  0.717510  1.898257 0.0576623  0.494248
Hsa-Let-7-P2a1_3p*                  18.6002      6.0987421  2.681041  2.274766 0.0229200  0.347014
Hsa-Let-7-P2a3_5p                10822.5933      0.0444091  0.395465  0.112296 0.9105888  0.990665
...                                     ...            ...       ...       ...       ...       ...
Hsa-Mir-95-P1_3p                     22.921      -3.512453  2.149234 -1.634281 0.1021999  0.655709
Hsa-Mir-95-P2_3p                    288.942       0.282295  0.789321  0.357643 0.7206107  0.955427
Hsa-Mir-96-P1_5p                    193.937       0.657772  0.815933  0.806159 0.4201511  0.951261
Hsa-Mir-96-P2_5p                   5249.788      -0.405491  0.405309 -1.000450 0.3170930  0.930944
Hsa-Mir-96-P3_5p                    205.622       2.158153  1.010105  2.136563 0.0326336  0.391603

MA plot

Фильтрация точек с низким средним экспрессированием (по baseMean). • Обычно отсекаются baseMean < 1. 2. Определение значимых генов (синие точки): • Используется критерий padj < 0.1 по умолчанию, а не < 0.05!

tiff("./pictures/PlotMA_standart_padj_0.05_type150.tiff", 
     width = 8, height = 6, units = "in", res = 300, bg = "white")
plotMA(res_150, alpha = 0.05, ylim = c(-8, 8)) 
dev.off()

null device 
          1 

plotMA(res_150, alpha = 0.05, ylim = c(-8, 8)) 

Кастомный MA plot по p-value

res_df <- res_150 %>%
  as.data.frame %>%
  mutate(color = case_when( 
    pvalue < 0.05  ~ "blue", # Значимые только по p-value
    TRUE ~ "gray70"
  ))

ggplot(res_df, aes(x = baseMean, y = log2FoldChange, color = color)) +
  geom_point(alpha = 0.7, size = 1) +
  geom_hline(yintercept = 0, linetype = "solid", color = "gray40", size = 1.5) +  # Добавляем линию
  scale_color_manual(values = c("gray70" = "gray70", "blue" = "blue")) +
  scale_x_log10(labels = scales::scientific) + 
  theme_minimal() +
  labs(x = "mean of normalized counts", y = "log fold change") +
  theme(legend.position = "none")

Значимые результаты

signres_150 <- results(dds_150, contrast=c("condition", "before", "after"), alpha=0.05) 
summary(signres_150)

out of 240 with nonzero total read count
adjusted p-value < 0.05
LFC > 0 (up)       : 1, 0.42%
LFC < 0 (down)     : 3, 1.2%
outliers [1]       : 0, 0%
low counts [2]     : 0, 0%
(mean count < 6)
[1] see 'cooksCutoff' argument of ?results
[2] see 'independentFiltering' argument of ?results

Let’s arranged it by log2FoldChange:

order_indices <- order(-res_150$log2FoldChange)
res_150[order_indices, ]

log2 fold change (MLE): condition before vs after 
Wald test p-value: condition before vs after 
DataFrame with 240 rows and 6 columns
                               baseMean log2FoldChange     lfcSE      stat      pvalue        padj
                              <numeric>      <numeric> <numeric> <numeric>   <numeric>   <numeric>
Hsa-Mir-150_3p*                 41.0827        6.48415   2.05853   3.14990 0.001633280  0.07839744
Hsa-Mir-328_3p                  92.7095        6.41844   1.64997   3.89004 0.000100228  0.00801828
Hsa-Mir-145_3p*                 44.0673        6.31701   2.05972   3.06693 0.002162671  0.08086155
Hsa-Let-7-P2a1_3p*              18.6002        6.09874   2.68104   2.27477 0.022919964  0.34701443
Hsa-Mir-33-P2_5p*               17.7651        6.08611   2.76046   2.20475 0.027471976  0.34701443
...                                 ...            ...       ...       ...         ...         ...
Hsa-Mir-133-P1_3p/P2_3p/P3_3p  778.6034       -4.27149  0.634382  -6.73331 1.65852e-11 1.99022e-09
Hsa-Mir-431_5p                  18.5437       -4.69940  2.595322  -1.81072 7.01841e-02 5.80834e-01
Hsa-Mir-219-P1_5p*              11.4359       -4.76811  3.574814  -1.33381 1.82267e-01 8.27845e-01
Hsa-Mir-499_5p                  68.1969       -5.58010  1.488728  -3.74824 1.78082e-04 1.06849e-02
Hsa-Mir-208-P2_3p              544.5306       -7.31750  1.032542  -7.08688 1.37167e-12 3.29202e-10

Visualisation for the first gene

plotCounts(dds_150, gene=which.max(res_150$log2FoldChange), intgroup="condition")

plotCounts(dds_150, gene=which.min(res_150$padj), intgroup="condition")

#plotCounts(dds, gene=rownames(res)[which.min(res$padj[which.max(res$log2FoldChange)])], intgroup="condition")

Volcano plot

plt <- EnhancedVolcano(res_150,
                lab = rownames(res_150),
                x = "log2FoldChange",
                y = "padj",
                pCutoff = 0.05,
                FCcutoff = 1,
                labSize = 3.0,
                boxedLabels = FALSE,
                col = c('black', '#CBD5E8', '#B3E2CD', '#FDCDAC'),
                colAlpha = 1,
                title = NULL,        
                subtitle = NULL) 

plt
ggsave("./pictures/Volcano plot_150type.tiff", plot = plt, width = 8, height = 6, dpi = 300, bg = "white")

rlog трансформация • Черные точки – стандартное отклонение отдельных генов. • Красная линия – сглаженный тренд зависимости SD от среднего значения экспрессии. • Если красная линия наклонена вверх → стандартное отклонение растёт с увеличением среднего (плохая нормализация). • Если красная линия примерно горизонтальна → нормализация сработала хорошо.

rlt_150 <- rlog(dds_150) 
meanSdPlot(assay(rlt_150))  #показывает, как изменяется стандартное отклонение в зависимости от среднего значения экспрессии.

PCA plot PCA – это метод снижения размерности, который показывает различия между образцами. Строится на основе различий в экспрессии по всем генам, даже если они не являются значимо дифференциально экспрессированными. • Оно не накладывает статистических порогов вроде padj < 0.05 или |log2FoldChange| > 1, а просто ищет наибольшие источники вариабельности среди всех измерений. • Поэтому даже небольшие изменения экспрессии могут формировать кластеры и различать группы в PCA, если эти изменения систематичны между образцами.

pcaData <- plotPCA(rlt_150, intgroup=c("condition", "patient"), returnData = TRUE)

using ntop=500 top features by variance

percentVar <- round(100 * attr(pcaData, "percentVar"))

ggplot(pcaData, aes(PC1, PC2, shape = patient, color = condition)) +
  geom_point(size = 3) +
  xlab(paste0("PC1: ", percentVar[1], "%")) +
  ylab(paste0("PC2: ", percentVar[2], "%")) + 
  coord_fixed() +
  theme_bw() +
  scale_color_brewer(palette = "Set2")

assay(rlt_150) <- limma::removeBatchEffect(assay(rlt_150),
                                       batch = colData(dds_150)[,'patient'])

pcaData <- plotPCA(rlt_150, intgroup=c("condition", "patient"), returnData = TRUE)

using ntop=500 top features by variance

percentVar <- round(100 * attr(pcaData, "percentVar"))

plt <- ggplot(pcaData, aes(PC1, PC2, shape = patient, color = condition)) +
  geom_point(size = 3) +
  xlab(paste0("PC1: ", percentVar[1], "%")) +
  ylab(paste0("PC2: ", percentVar[2], "%")) + 
  coord_fixed() +
  theme_bw() +
  scale_color_brewer(palette = "Set2")
plt
ggsave("./pictures/PCA plot for type 150 after removing donor effect.tiff", plot = plt, width = 8, height = 6, dpi = 300, bg = "white")

Plot a heatmap of diff expressed genes

res_sign_150 <- subset(res_150, padj < 0.05 & !is.na(padj) & abs(log2FoldChange) > 1.0)
res_sign_150 <- res_sign_150[order(res_sign_150$log2FoldChange, decreasing = TRUE), ]

sig_genes <- rownames(res_sign_150)  # Получаем имена генов, которые прошли фильтрацию

de_mat <- assay(rlt_150)[sig_genes, ] 
datamatrix <- t(scale(t(de_mat)))

annotation_col <- data.frame(condition = coldata_150$condition)
rownames(annotation_col) <- colnames(datamatrix)

annotation_colors <- list(
  condition = c("before" = "#FFCC00", "after" = "#3399FF")
)

plt <- pheatmap(datamatrix, 
         cluster_rows = TRUE, 
         show_rownames = TRUE, 
         cluster_cols = TRUE, 
         annotation_col = annotation_col,
         annotation_colors = annotation_colors,
         display_numbers = TRUE,
         legend = FALSE,
         fontsize = 15)  
plt

ggsave("./pictures/Heatmap of diff expressed genes_type150.tiff", plot = plt, width = 8, height = 6, dpi = 300, bg = "white")

Plot of the distance between samples heatmap Расчет расстояний между образцами • Обычно используется евклидово расстояние (по умолчанию в DESeq2). • Оно вычисляется по нормализованным данным экспрессии (rlog() или vst()). • Чем меньше расстояние — тем более похожи образцы.

sampleDists_150 <- dist(t(assay(rlt_150)))
sampleDistMatrix_150 <- as.matrix(sampleDists_150)
rownames(sampleDistMatrix_150) <- paste(rlt_150$condition, rlt_150$patient, sep="_patient")
colnames(sampleDistMatrix_150) <- paste(rlt_150$condition, rlt_150$patient, sep="_patient")
colors <- colorRampPalette(rev(brewer.pal(9, "Blues")) )(255)

plt <- pheatmap(sampleDistMatrix_150,
         clustering_distance_rows = "euclidean",
         clustering_distance_cols = "euclidean",
         fontsize = 12,
         legend = FALSE,
         display_numbers = TRUE,
         color = colors)
plt

ggsave("./pictures/Plot of the distance between samples_type150.tiff", plot = plt, width = 8, height = 6, dpi = 300, bg = "white")

Анализ обогащения для везикул типа 150

up_150 <- res_sign_150 %>% 
  as.data.frame() %>% 
  filter(log2FoldChange > 0)
down_150 <- res_sign_150 %>% 
  as.data.frame() %>% 
  filter(log2FoldChange < 0)
rownames(up_150)

[1] "Hsa-Mir-328_3p"

rownames(down_150)

[1] "Hsa-Mir-133-P1_3p/P2_3p/P3_3p" "Hsa-Mir-499_5p"                "Hsa-Mir-208-P2_3p"            

перевела вручную: https://mirgenedb.org/browse/hsa Hsa-Mir-328 = hsa-mir-328 Eutheria — клада зверей, включающая плацентарных и различные вымершие инфраклассы. Hsa-Mir-133-P1 = hsa-mir-133a-1 Gnathostomata Hsa-Mir-133-P2 = hsa-mir-133b Gnathostomata Hsa-Mir-133-P3 = hsa-mir-133a-2 Gnathostomata есть только hsa-miR-133a-3p Hsa-Mir-499 = hsa-mir-499a Vertebrata Hsa-Mir-208-P2 hsa-mir-208a Vertebrata • miRBase: https://www.mirbase.org/ • MirGeneDB: https://mirgenedb.org/

mirna_names_up <- c("hsa-miR-328-3p")
mirna_names_down <- c("hsa-miR-133a-3p", "hsa-miR-133a-3p", "hsa-miR-208a-3p", "hsa-miR-499a-5p")

Конвертация в MIMATID

converted_mirna_up <- miRNAVersionConvert(mirna_names_up)
converted_mirna_down <- miRNAVersionConvert(mirna_names_down)
converted_mirna_up

converted_mirna_down

Запрос таргетов из базы multiMiR

targets150_down <- unique(get_multimir(org = "hsa", mirna = converted_mirna_up$Accession, table = "validated")@data$target_symbol)

Searching mirecords ...
Searching mirtarbase ...
Searching tarbase ...

targets150_up <- unique(get_multimir(org = "hsa", mirna = converted_mirna_down$Accession, table = "validated")@data$target_symbol)

Searching mirecords ...
Searching mirtarbase ...
Searching tarbase ...

#writeLines(targets_down, "targets_down150_list.txt")
#writeLines(targets_up, "targets_up150_list.txt")

Анализ обогащения из базы биологических процессов

#msig_go_bp <- msigdbr(species = "Homo sapiens", category = "C5", subcategory = "GO:BP")
# targets_down <- readLines("targets_down150_list.txt")
# targets_up <- readLines("targets_up150_list.txt")

GO_enrich_down150_bp <- enrichGO(
  gene          = targets150_down,  
  OrgDb         = org.Hs.eg.db,
  keyType       = "SYMBOL",
  ont           = "BP", 
  pAdjustMethod = "BH",
  qvalueCutoff  = 0.05
)

GO_enrich_up150_bp <- enrichGO(
  gene          = targets150_up,  
  OrgDb         = org.Hs.eg.db,
  keyType       = "SYMBOL",
  ont           = "BP", 
  pAdjustMethod = "BH",
  qvalueCutoff  = 0.05
)

Визуализация

p1 <- dotplot(GO_enrich_down150_bp, showCategory = 20) + 
  ggtitle("GO Enrichment for DOWNregulated targets") + 
  theme(
    plot.title = element_text(size = 12, face = "bold"),
    axis.text.y = element_text(size = 11)
  )

p2 <- dotplot(GO_enrich_up150_bp, showCategory = 20) + 
  ggtitle("GO Enrichment for UPregulated targets") + 
  theme(
    plot.title = element_text(size = 12, face = "bold"),
    axis.text.y = element_text(size = 11)
  )

p1 + p2  

combined_plot <- p1 + p2

ggsave("./pictures/GO_enrichment_dotplot_BP_type150.tiff", plot = combined_plot, width = 16, height = 10, dpi = 300)

GO_enrich_DOWN150_BP <- enrichplot::pairwise_termsim(GO_enrich_down150_bp, method = "JC")

plt <- emapplot(GO_enrich_DOWN150_BP, 
         repel = TRUE,
         showCategory = 20) +
  ggtitle("Biological processes for DOWNregulated targets for vesicles 150") +
  theme(
    plot.title = element_text(size = 12, face = "bold"),
    axis.text = element_text(size = 3)
    )    

plt

ggsave("./pictures/GO_enrichment_emapplot_BPdown_type150.tiff", plot = plt, width = 16, height = 10, dpi = 300)

GO_enrich_UP150_BP <- enrichplot::pairwise_termsim(GO_enrich_up150_bp, method = "JC")

plt <- emapplot(GO_enrich_UP150_BP, 
         repel = TRUE,
         showCategory = 20) +
  ggtitle("Biological processes for UPregulated targets for vesicles 150") +
  theme(
    plot.title = element_text(size = 12, face = "bold"),
    axis.text = element_text(size = 3)
    )   

plt

ggsave("./pictures/GO_enrichment_emapplot_BPup_type150.tiff", plot = plt, width = 16, height = 10, dpi = 300)

Анализ обогащения из базы IMMUNESIGDB IMMUNESIGDB: Наборы генов, связанные с иммунной системой, включая иммуно-онкологию и другие аспекты иммунитета.

msig_go_bp <- msigdbr(species = "Homo sapiens", category = "C7", subcategory = "IMMUNESIGDB")

GO_enrich_up150_immum <- enricher(gene = targets150_up, TERM2GENE = msig_go_bp[, c("gs_name", "gene_symbol")])
GO_enrich_down150_immun <- enricher(gene = targets150_down, TERM2GENE = msig_go_bp[, c("gs_name", "gene_symbol")])

Визуализация IMMUNESIGDB

p1 <- dotplot(GO_enrich_up150_immum, showCategory = 20) + 
  ggtitle("GO Enrichment IMMUNESIGD for UPregulated targets for vesicles 150") + 
  theme(
    plot.title = element_text(size = 12, face = "bold"),
    axis.text.y = element_text(size = 8)
  )

p2 <- dotplot(GO_enrich_down150_immun, showCategory = 20) + 
  ggtitle("GO Enrichment IMMUNESIGD for DOWNregulated targets for vesicles 150") + 
  theme(
    plot.title = element_text(size = 12, face = "bold"),
    axis.text.y = element_text(size = 8)
  )

p1 + p2

combined_plot <- p1 + p2

ggsave("./pictures/GO_enrichment_dotplot_IMMUNESIGD_type150.tiff", plot = combined_plot, width = 16, height = 10, dpi = 300)

Анализ обогащения из базы KEGG

msig_go_bp <- msigdbr(species = "Homo sapiens", category = "C2", subcategory = "CP:KEGG")

GO_enrich_up150_KEGG <- enricher(gene = targets150_up, TERM2GENE = msig_go_bp[, c("gs_name", "gene_symbol")])
GO_enrich_down150_KEGG <- enricher(gene = targets150_down, TERM2GENE = msig_go_bp[, c("gs_name", "gene_symbol")])

p1 <- dotplot(GO_enrich_up150_KEGG, showCategory = 20) + 
  ggtitle("GO Enrichment KEGG for UPregulated targets for vesicles 150") + 
  theme(
    plot.title = element_text(size = 12, face = "bold"),
    axis.text.y = element_text(size = 8)
  )

p2 <- dotplot(GO_enrich_down150_KEGG, showCategory = 20) + 
  ggtitle("GO Enrichment KEGG for DOWNregulated targets for vesicles 150") + 
  theme(
    plot.title = element_text(size = 12, face = "bold"),
    axis.text.y = element_text(size = 8)
  )

p1 + p2

combined_plot <- p1 + p2

ggsave("./pictures/GO_enrichment_dotplot_KEGG_type150.tiff", plot = combined_plot, width = 16, height = 10, dpi = 300)

Анализ обогащения из базы CP:WIKIPATHWAYS

msig_go_bp <- msigdbr(species = "Homo sapiens", category = "C2", subcategory = "CP:WIKIPATHWAYS")

GO_enrich_up150_WIKI <- enricher(gene = targets150_up, TERM2GENE = msig_go_bp[, c("gs_name", "gene_symbol")])
GO_enrich_down150_WIKI <- enricher(gene = targets150_down, TERM2GENE = msig_go_bp[, c("gs_name", "gene_symbol")])

p1 <- dotplot(GO_enrich_up150_WIKI, showCategory = 20) + 
  ggtitle("GO Enrichment WIKIPATHWAYS for UPregulated targets for vesicles 150") + 
  theme(
    plot.title = element_text(size = 12, face = "bold"),
    axis.text.y = element_text(size = 8)
  )

p2 <- dotplot(GO_enrich_down150_WIKI, showCategory = 20) + 
  ggtitle("GO Enrichment WIKIPATHWAYS for DOWNregulated targets for vesicles 150") + 
  theme(
    plot.title = element_text(size = 12, face = "bold"),
    axis.text.y = element_text(size = 8)
  )

combined_plot <- p1 + p2

combined_plot

ggsave("./pictures/GO_enrichment_dotplot_WIKIPATHWAY_type150.tiff", plot = combined_plot, width = 16, height = 10, dpi = 300)

Только везикулы типа 16

coldata_16 <- coldata[coldata$type == 16, ]
rownames(coldata_16) <- coldata_16$sample
coldata_16

common_samples_16 <- intersect(colnames(counts), coldata_16$samples)

counts_16 <- counts[, c(counts$miRNA, common_samples)]  
counts_16 <- counts_16[, rownames(coldata_16)] #ранжирую по колонки в counts так же как и названия строк в coldata_150
head(counts_16)

Создаем DESeqDataSet из матрицы каунтов

dds_16 <- DESeqDataSetFromMatrix(countData = counts_16, 
                              colData = coldata_16, 
                              design = ~ 0 + patient + condition)

converting counts to integer mode

dds_16$condition <- relevel(dds_16$condition, ref = "before")
dds_16

class: DESeqDataSet 
dim: 913 6 
metadata(1): version
assays(1): counts
rownames(913): Hsa-Let-7-P1a_3p* Hsa-Let-7-P1a_5p/P2a1_5p/P2a2_5p ... Hsa-Mir-9851_3p Hsa-Mir-9851_5p*
rowData names(0):
colnames(6): 29.1p16_S39_R1_001 15.1p16_S33_R1_001 ... 15.7p16_S31_R1_001 17.7p16_S37_R1_001
colData names(4): sample condition type patient

Фильтрация

dim(dds_16)

[1] 913   6

smallestGroupSize <- 3
keep <- rowSums(counts(dds_16) >= 10) >= smallestGroupSize
dds_16 <- dds_16[keep,]
dim(dds_16)

[1] 219   6

Run Differential Expression Analysis for 150 type

dds_16 <- DESeq(dds_16, fitType = "parametric")

using pre-existing size factors
estimating dispersions
found already estimated dispersions, replacing these
gene-wise dispersion estimates
mean-dispersion relationship
final dispersion estimates
fitting model and testing

plotDispEsts(dds_16)

res_16 <- results(dds_16, contrast=c("condition", "before", "after"))
res_16

log2 fold change (MLE): condition before vs after 
Wald test p-value: condition before vs after 
DataFrame with 219 rows and 6 columns
                                   baseMean log2FoldChange     lfcSE      stat    pvalue      padj
                                  <numeric>      <numeric> <numeric> <numeric> <numeric> <numeric>
Hsa-Let-7-P1a_5p/P2a1_5p/P2a2_5p 22993.7673      0.1901164  0.224536  0.846707 0.3971586  0.926559
Hsa-Let-7-P1b_5p                  2527.3696      0.4961225  0.280481  1.768830 0.0769222  0.575168
Hsa-Let-7-P1c_5p                   481.2148      0.5330152  0.461003  1.156209 0.2475958  0.860794
Hsa-Let-7-P2a1_3p*                  23.7282      0.4135740  1.966411  0.210319 0.8334185  0.954993
Hsa-Let-7-P2a3_5p                45006.2426     -0.0642043  0.238029 -0.269733 0.7873659  0.940503
...                                     ...            ...       ...       ...       ...       ...
Hsa-Mir-92-P2c_5p*                  25.3899      -0.754468  2.275831 -0.331513 0.7402570  0.931703
Hsa-Mir-95-P2_3p                   193.6126      -0.340069  0.662909 -0.512995 0.6079550  0.926559
Hsa-Mir-96-P1_5p                    14.4204      -4.216533  2.813830 -1.498503 0.1340025  0.764653
Hsa-Mir-96-P2_5p                  4249.9579      -0.537344  0.268063 -2.004539 0.0450123  0.487244
Hsa-Mir-96-P3_5p                   121.7907       0.504262  0.806362  0.625355 0.5317382  0.926559

MA plot

Фильтрация точек с низким средним экспрессированием (по baseMean). • Обычно отсекаются baseMean < 1. 2. Определение значимых генов (синие точки): • Используется критерий padj < 0.1 по умолчанию, а не < 0.05!

tiff("./pictures/PlotMA_standart_padj_0.05_type16.tiff", 
     width = 8, height = 6, units = "in", res = 300, bg = "white")
plotMA(res_16, alpha = 0.05, ylim = c(-8, 8)) 
dev.off()

null device 
          1 

plotMA(res_16, alpha = 0.05, ylim = c(-8, 8)) 

Кастомный MA plot по p-value

res_df <- res_16 %>%
  as.data.frame %>%
  mutate(color = case_when( 
    pvalue < 0.05 & !is.na(pvalue) & abs(log2FoldChange) > 1  ~ "blue",  # Значимые по p-value и диф экспрессированные
    TRUE ~ "gray70"
  ))

plt <- ggplot(res_df, aes(x = baseMean, y = log2FoldChange, color = color)) +
  geom_point(alpha = 0.7, size = 1) +
  geom_hline(yintercept = 0, linetype = "solid", color = "gray40", size = 1.5) +  # Добавляем линию
  scale_color_manual(values = c("gray70" = "gray70", "blue" = "blue")) +
  scale_x_log10(labels = scales::scientific) + 
  theme_minimal() +
  labs(x = "mean of normalized counts", y = "log fold change") +
  theme(legend.position = "none")

plt
ggsave("./pictures/PlotMA_castom_pvalue0.05_type16.tiff", plot = plt, width = 8, height = 6, dpi = 300, bg = "white")

Значимые результаты

summary(results(dds_16, contrast=c("condition", "before", "after"), alpha=0.05))

out of 219 with nonzero total read count
adjusted p-value < 0.05
LFC > 0 (up)       : 0, 0%
LFC < 0 (down)     : 0, 0%
outliers [1]       : 0, 0%
low counts [2]     : 0, 0%
(mean count < 8)
[1] see 'cooksCutoff' argument of ?results
[2] see 'independentFiltering' argument of ?results

Let’s arranged it by log2FoldChange:

order_indices <- order(-res_16$log2FoldChange)
res_16[order_indices, ]

log2 fold change (MLE): condition before vs after 
Wald test p-value: condition before vs after 
DataFrame with 219 rows and 6 columns
                    baseMean log2FoldChange     lfcSE      stat     pvalue      padj
                   <numeric>      <numeric> <numeric> <numeric>  <numeric> <numeric>
Hsa-Mir-185_5p       14.8310        5.85950   2.69426   2.17481 0.02964445  0.487244
Hsa-Mir-136_5p*      13.3826        5.83711   3.16104   1.84658 0.06480773  0.575168
Hsa-Mir-197_3p       45.5597        4.16951   1.59282   2.61769 0.00885267  0.357073
Hsa-Mir-101-P1_5p*   16.5354        4.08972   2.30004   1.77811 0.07538608  0.575168
Hsa-Mir-10-P3a_5p    37.9479        3.82488   1.69051   2.26256 0.02366292  0.487244
...                      ...            ...       ...       ...        ...       ...
Hsa-Mir-154-P17_3p   15.3729       -4.07531   2.60884  -1.56211  0.1182617  0.764653
Hsa-Mir-190-P1_5p    10.5910       -4.10593   3.44201  -1.19289  0.2329137  0.850135
Hsa-Mir-96-P1_5p     14.4204       -4.21653   2.81383  -1.49850  0.1340025  0.764653
Hsa-Mir-128-P1_5p*   15.3208       -4.49032   3.05489  -1.46988  0.1415942  0.764653
Hsa-Mir-197_5p*      17.0795       -5.15605   2.93332  -1.75775  0.0787901  0.575168

Visualisation for the first gene

plotCounts(dds_16, gene=which.max(res_16$log2FoldChange), intgroup="condition")

plotCounts(dds_16, gene=which.min(res_16$pvalue), intgroup="condition")

#plotCounts(dds, gene=rownames(res)[which.min(res$padj[which.max(res$log2FoldChange)])], intgroup="condition")

Volcano plot

plt <- EnhancedVolcano(res_16,
                lab = rownames(res_16),
                x = "log2FoldChange",
                y = "pvalue",
                pCutoff = 0.05,
                FCcutoff = 1,
                labSize = 3.0,
                boxedLabels = FALSE,
                col = c('black', '#CBD5E8', '#B3E2CD', '#FDCDAC'),
                colAlpha = 1,
                title = NULL,        
                subtitle = NULL) 

plt
ggsave("./pictures/Volcano plot_based_on_Pvalue_16type.tiff", plot = plt, width = 8, height = 6, dpi = 300, bg = "white")

rlog трансформация • Черные точки – стандартное отклонение отдельных генов. • Красная линия – сглаженный тренд зависимости SD от среднего значения экспрессии. • Если красная линия наклонена вверх → стандартное отклонение растёт с увеличением среднего (плохая нормализация). • Если красная линия примерно горизонтальна → нормализация сработала хорошо.

rlt_16 <- rlog(dds_16) 
meanSdPlot(assay(rlt_16))  #показывает, как изменяется стандартное отклонение в зависимости от среднего значения экспрессии.

pcaData <- plotPCA(rlt_16, intgroup=c("condition", "patient"), returnData = TRUE)

using ntop=500 top features by variance

percentVar <- round(100 * attr(pcaData, "percentVar"))

ggplot(pcaData, aes(PC1, PC2, shape = patient, color = condition)) +
  geom_point(size = 3) +
  xlab(paste0("PC1: ", percentVar[1], "%")) +
  ylab(paste0("PC2: ", percentVar[2], "%")) + 
  coord_fixed() +
  theme_bw() +
  scale_color_brewer(palette = "Set2")

assay(rlt_16) <- limma::removeBatchEffect(assay(rlt_16),
                                       batch = colData(dds_16)[,'patient'])

pcaData <- plotPCA(rlt_16, intgroup=c("condition", "patient"), returnData = TRUE)

using ntop=500 top features by variance

percentVar <- round(100 * attr(pcaData, "percentVar"))

plt <- ggplot(pcaData, aes(PC1, PC2, shape = patient, color = condition)) +
  geom_point(size = 3) +
  xlab(paste0("PC1: ", percentVar[1], "%")) +
  ylab(paste0("PC2: ", percentVar[2], "%")) + 
  coord_fixed() +
  theme_bw() +
  scale_color_brewer(palette = "Set2")

plt
ggsave("./pictures/PCA plot for type 16 after removing donor effect.tiff", plot = plt, width = 8, height = 6, dpi = 300, bg = "white")

Plot a heatmap of the most expressed genes

res_sign_16 <- subset(res_16, pvalue < 0.05 & !is.na(pvalue) & abs(log2FoldChange) > 1.0)
res_sign_16 <- res_sign_16[order(res_sign_16$log2FoldChange, decreasing = TRUE), ]

sig_genes <- rownames(res_sign_16)  # Получаем имена генов, которые прошли фильтрацию

de_mat <- assay(rlt_16)[sig_genes, ] 
datamatrix <- t(scale(t(de_mat)))

annotation_col <- data.frame(condition = coldata_16$condition)
rownames(annotation_col) <- colnames(datamatrix)

annotation_colors <- list(
  condition = c("before" = "#FFCC00", "after" = "#3399FF")
)

plt <- pheatmap(datamatrix, 
         cluster_rows = TRUE, 
         show_rownames = TRUE, 
         cluster_cols = TRUE, 
         annotation_col = annotation_col,
         annotation_colors = annotation_colors,
         display_numbers = TRUE,
         legend = FALSE,
         fontsize = 15)  

plt

ggsave("./pictures/Heatmap of diff expressed genes_type16.tiff", plot = plt, width = 8, height = 6, dpi = 300, bg = "white")

Plot of the distance between samples heatmap Расчет расстояний между образцами • Обычно используется евклидово расстояние (по умолчанию в DESeq2). • Оно вычисляется по нормализованным данным экспрессии (rlog() или vst()). • Чем меньше расстояние — тем более похожи образцы.

sampleDists_16 <- dist(t(assay(rlt_16)))
sampleDistMatrix_16 <- as.matrix(sampleDists_16)
rownames(sampleDistMatrix_16) <- paste(rlt_16$condition, rlt_16$patient, sep="_patient")
colnames(sampleDistMatrix_16) <- paste(rlt_16$condition, rlt_16$patient, sep="_patient")

plt <- pheatmap(sampleDistMatrix_16,
         clustering_distance_rows = "euclidean",
         clustering_distance_cols = "euclidean",
         fontsize = 12,
         legend = FALSE,
         display_numbers = TRUE,
         color = colors)

plt

ggsave("./pictures/Plot of the distance between samples_type16.tiff", plot = plt, width = 8, height = 6, dpi = 300, bg = "white")

Анализ обогащения для везикул типа 16

up_16 <- res_sign_16 %>% 
  as.data.frame() %>% 
  filter(log2FoldChange > 0)
down_16 <- res_sign_16 %>% 
  as.data.frame() %>% 
  filter(log2FoldChange < 0)
rownames(up_16)

[1] "Hsa-Mir-185_5p"     "Hsa-Mir-197_3p"     "Hsa-Mir-10-P3a_5p"  "Hsa-Mir-148-P3_5p*" "Hsa-Mir-30-P1a_3p*" "Hsa-Mir-150_3p*"   
[7] "Hsa-Mir-361_3p*"    "Hsa-Mir-10-P2b_5p"  "Hsa-Mir-126_3p*"   

rownames(down_16)

[1] "Hsa-Mir-340_5p"     "Hsa-Mir-181-P2c_5p" "Hsa-Mir-223_5p*"    "Hsa-Mir-30-P1c_3p*"

Переводим в miRBase • miRBase: https://www.mirbase.org/ • MirGeneDB: https://mirgenedb.org/

url <- "https://mirgenedb.org/browse/hsa"
page <- read_html(url)

Парсим таблицу

mir_table <- page %>%
  html_element("table") %>%
  html_table(fill = TRUE) 

mir_table <- mir_table[-c(1:3), c(1,2) ] 
colnames(mir_table) <- c("MirGeneDB_ID", "MiRBase_ID")
mir_table$MirGeneDB_ID <- sub(" V", "", mir_table$MirGeneDB_ID)

head(mir_table)

up_16_clean <- sub("_.*", "", row.names(up_16))
up_16_converted <- mir_table$MiRBase_ID[match(up_16_clean, mir_table$MirGeneDB_ID)]

down_16_clean <- sub("_.*", "", row.names(down_16))
down_16_converted <- mir_table$MiRBase_ID[match(down_16_clean, mir_table$MirGeneDB_ID)]
up_16_converted

[1] "hsa-mir-185" "hsa-mir-197" NA            "hsa-mir-152" "hsa-mir-30d" "hsa-mir-150" "hsa-mir-361" "hsa-mir-99b" NA           

down_16_converted

[1] "hsa-mir-340"  "hsa-mir-181d" "hsa-mir-223"  NA            

Конвертация в MIMATID в итоге заменила NA вручную на самые близкие, но это такая себе практика

NA Hsa-Mir-10-P3a_5p есть два соответствия: Hsa-Mir-10-P1c = hsa-mir-10a Hsa-Mir-10-P3b = hsa-mir-125a

NA Hsa-Mir-126_3p* есть одно соответствие: Hsa-Mir-126-P2 = hsa-mir-126

NA Hsa-Mir-30-P1c_3p* есть три соответствия: Hsa-Mir-30-P1a = hsa-mir-30d Hsa-Mir-30-P1b = hsa-mir-30a Hsa-Mir-30-P1d = hsa-mir-30e

[1] “Hsa-Mir-185_5p” “Hsa-Mir-197_3p” “Hsa-Mir-10-P3a_5p” “Hsa-Mir-148-P3_5p” ”Hsa-Mir-30-P1a_3p” “Hsa-Mir-150_3p”
[7] ”Hsa-Mir-361_3p” “Hsa-Mir-10-P2b_5p” “Hsa-Mir-126_3p”
[1] ”Hsa-Mir-340_5p” ”Hsa-Mir-181-P2c_5p” ”Hsa-Mir-223_5p” “Hsa-Mir-30-P1c_3p*”

MI (MicroRNA Gene ID) — это идентификатор предшественника (precursor) miRNA MIMAT (Mature miRNA ID) — это идентификатор зрелой (mature) miRNA, которая функционирует в клетке

up_16_converted <-  c("hsa-mir-185-5p", "hsa-mir-197-3p", "hsa-mir-152-5p", "hsa-mir-30d-3p", "hsa-mir-150-3p", "hsa-mir-361-3p", "hsa-mir-99b-5p", "hsa-mir-126-3p")
down_16_converted <- c("hsa-mir-340-5p", "hsa-mir-181d-5p", "hsa-mir-223-5p")

# up_16_converted <-  c("hsa-mir-185", "hsa-mir-197", "hsa-mir-152", "hsa-mir-30d", "hsa-mir-150", "hsa-mir-361", "hsa-mir-99b", "hsa-mir-126")
# down_16_converted <- c("hsa-mir-340", "hsa-mir-181d", "hsa-mir-223") #но есть только MI, а не MIMAT

converted_mirna_up16 <- miRNAVersionConvert(up_16_converted)
converted_mirna_down16 <- miRNAVersionConvert(down_16_converted)
converted_mirna_up16

converted_mirna_down16

Запрос таргетов из базы multiMiR

# targets16_up <- unique(get_multimir(org = "hsa", mirna = converted_mirna_up16$Accession, table = "predicted")@data$target_symbol)
# targets16_up

# targets16_down <- unique(get_multimir(org = "hsa", mirna = converted_mirna_down16$Accession, table = "predicted")@data$target_symbol)
# targets16_down